First-party vs third-party import detection for modules built with C++/Rust #17761
Description
Summary
This is a spin out of #16712
Consider this setup:
Consider the following setup
$ tree
.
├── crates
│ └── my-package
│ └── pyproject.toml
├── known.py
├── main.py
└── pyproject.toml
3 directories, 4 files
With these contents:
$ cat crates/my-package/pyproject.toml
[project]
name = "my-package"
version = "0.1.0"
[build-system]
build-backend = "maturin"
requires = ["maturin>=1.0,<2.0"]
[tool.maturin]
bindings = "pyo3"
module-name = "my_package"
$ cat known.py
$ cat main.py
import os
import sys
import my_package
import torch
import unknown
import known
def main():
print("Hello from myexample!")
sys / my_package / os / unknown / torch / known
if __name__ == "__main__":
main()
$ cat pyproject.toml
[project]
name = "myexample"
version = "0.1.0"
description = "Add your description here"
requires-python = ">=3.13"
dependencies = []
[tool.ruff]
src = [
".",
"crates/*",
]
[tool.ruff.lint]
preview = true
select = [
"I00", # unsorted-imports, missing-required-import.
]
(imagine there's also crates/my-package/src/lib.rs that implements a PyO3 package, say).
The sort order here is what ruff will produce:
import os
import sys
import my_package
import torch
import unknown
import known
However, my_package is first-party. The actual module will be produced as an .so upon building.
Note that having crates/my-package/my_package.pyi does not change the behavior.
There seems to be no setting for src in [tool.ruff] that makes it identify the my_package import as first-party.
Initially I assumed it was the underscore that mislead it, but renaming crates/my-package to crates/my_package does not change its behavior.
Version
ruff 0.11.7